The Transfer Operator

From Dynamics to Spectrum

The Collatz map acts on individual numbers, but it induces a linear operator on the space of distributions over $\mathbb{Z}/M\mathbb{Z}$. This is the Perron-Frobenius (or Ruelle) transfer operator:

$$\mathcal{L}[g](y) = \sum_{f(x) = y} \frac{g(x)}{|f'(x)|}$$

For the Collatz map, the two branches contribute differently:

• Even branch $x \mapsto x/2$: derivative $1/2$, weight $2$
• Odd branch $x \mapsto 3x+1$: derivative $3$, weight $1/3$

Multiplier n: 3 Modulus M: 612244896

Transfer operator for 3x+1, x/2 on Z/24Z

Eigenvalue Spectrum

Spectral Analysis

Dimension	24
Non-zero eigenvalues	4
λ₁ (trivial)	2.0000
Critical radius |λ₂|	1.100642
Predicted radius	(2²/3)^(1/3) = 1.100642
|λ₂|ⁿ	1.333333
y²/n = 4/3	1.333333
Eigenvalue eq.	λ3 = 1.3333

λ³ = 4/3
λ = ∛(4/3) · ωᵏ
ω = e^(2πi/3), k = 0, 1, 2

Critical circle radius > 1
⟹ Subcritical: orbits contract

The Spectrum

Computing the eigenvalues of $\mathcal{L}$ on $\mathbb{Z}/M\mathbb{Z}$ for any multiple of 6 reveals a striking fact: the operator has exactly four non-zero eigenvalues, regardless of dimension.

Eigenvalue	Value	Magnitude
$\lambda_1$	$2$	$2$
$\lambda_2$	$(4/3)^{1/3}$	$\approx 1.1006$
$\lambda_3$	$(4/3)^{1/3} \cdot \omega$	$\approx 1.1006$
$\lambda_4$	$(4/3)^{1/3} \cdot \omega^2$	$\approx 1.1006$

where $\omega = e^{2\pi i/3}$ is a primitive cube root of unity.

The operator has rank 4. All remaining eigenvalues are exactly zero. The entire dynamics of the Collatz map, at the operator level, is captured by four modes.

Why Cube Roots of 4/3?

The characteristic equation for the non-trivial eigenvalues is:

$$\lambda^3 = \frac{4}{3} = \frac{y^2}{n}$$

This arises because:

1. The exponent 3 comes from the 3-fold symmetry: the map $3x+1$ cycles through residue classes mod 3 (since $3x + 1 \equiv 1 \pmod{3}$ for all $x$). The operator decomposes into three sectors related by $\omega$.

2. The ratio $4/3$ is the energy balance: each halving contributes weight $y = 2$ (two inverse images), while each odd step contributes weight $1/n = 1/3$. Over one full cycle through all three sectors: $y^2/n = 4/3$.

For a general $(n, y)$ system, the eigenvalue equation becomes $\lambda^n = y^2/n$, and convergence requires $|y^2/n| > 1$, i.e., $y^2 > n$.

The Critical Circle

The non-trivial eigenvalues lie on a circle in the complex plane:

$$|\lambda| = \left(\frac{4}{3}\right)^{1/3} \approx 1.1006$$

This is the critical circle — the Collatz analogue of the critical line $\text{Re}(s) = 1/2$ in the theory of the Riemann zeta function.

The Hilbert-Polya conjecture proposes that the zeros of $\zeta(s)$ are eigenvalues of a self-adjoint operator, which would force them onto the critical line. Here, the eigenvalues of the Collatz transfer operator are forced onto the critical circle by the 3-fold rotational symmetry of the map.

Riemann Zeta	Collatz Transfer
Zeros on the critical line $\text{Re}(s) = 1/2$	Eigenvalues on the critical circle $|\lambda| = (4/3)^{1/3}$
Constrained by functional equation $\zeta(s) = \chi(s)\zeta(1-s)$	Constrained by 3-fold symmetry $\lambda \mapsto \omega\lambda$
Line position forced by self-adjointness	Circle position forced by rotational symmetry
$1/2$ from the balance of $\Gamma$ factors	$(4/3)^{1/3}$ from the balance of weights $y^2/n$

The Convergence Criterion

The trivial eigenvalue $\lambda_1 = 2$ controls the gross scaling (it comes from the halving map's weight). The dynamically relevant eigenvalues are $\lambda_2, \lambda_3, \lambda_4$ on the critical circle.

Convergence of all orbits requires:

$$\left(\frac{4}{3}\right)^{1/3} > 1 \quad \iff \quad \frac{4}{3} > 1 \quad \iff \quad 4 > 3$$

The entire Collatz conjecture, at the spectral level, reduces to the statement that 4 is greater than 3.

For $5x+1$: the eigenvalue equation gives $\lambda^5 = 4/5$, so the critical circle has radius $(4/5)^{1/5} \approx 0.955 < 1$. This means the transfer operator's non-trivial modes decay — the system mixes efficiently. But mixing alone doesn't imply convergence: the trivial eigenvalue $\lambda_1 = 2$ still dominates, and the energy input per kick ($\log_2 5 \approx 2.32$ bits) exceeds the average drain (2 bits). The spectral gap measures how fast the system forgets its initial distribution; the criticality $\mu$ measures whether the average orbit contracts. For $5x+1$, the system mixes well but grows on average — uniform divergence.

Self-Adjointness

The transfer operator $\mathcal{L}$ is not symmetric (it has asymmetry ratio $\approx 1.41$). However, its non-trivial spectrum is entirely real or comes in conjugate pairs with equal magnitude — exactly the behavior of a normal operator (one that commutes with its adjoint).

The 3-fold symmetry $\lambda \mapsto \omega\lambda$ is a stronger constraint than self-adjointness. It forces the eigenvalues onto a circle rather than a line, and the radius of that circle is determined by a single number: $y^2/n$.

The Berry-Keating Connection

The Berry-Keating program seeks an operator related to the classical Hamiltonian $H = xp$ whose eigenvalues give the Riemann zeros. In the Collatz setting, the analogous Hamiltonian is:

$$H_{\text{Collatz}} = \log_2(x) \cdot v_2(3x+1)$$

This is the product of "position" (the bit-length $\log_2 x$) and "momentum" (the 2-adic valuation $v_2(3x+1)$, which controls how many halvings follow each kick). The conservation law

$$s \cdot \log_2(6) = T - \log_2(n) + \varepsilon$$

is the Collatz analogue of $E = H(x, p)$ — the total energy expressed in terms of position and momentum.

Across the Zoo

The spectral structure changes predictably across the Collatz Zoo:

System	Critical circle radius	Eigenvalue equation	Convergent?
$3x+1, \; x/2$	$(4/3)^{1/3} \approx 1.10$	$\lambda^3 = 4/3$	Yes
$5x+1, \; x/2$	$(4/5)^{1/5} \approx 0.96$	$\lambda^5 = 4/5$	No
$7x+1, \; x/2$	$(4/7)^{1/7} \approx 0.92$	$\lambda^7 = 4/7$	No

The critical circle shrinks below 1 as $n$ exceeds $y^2 = 4$. Only $n = 3$ keeps the radius above 1.

Related

• Eisenstein Lattice — the eigenvalue equation as a ratio of Eisenstein norms
• Universal Dynamics — the thermodynamic framework and Collatz Zoo
• abc Conjecture — why $\text{rad}(6) = 6$ is the minimal radical
• 3-Adic Mixing — spectral gap on the Markov chain (related but distinct operator)
• The Hidden Rotation — the near-conjugacy to irrational rotation